使用無響室錄音合成虛擬聆聽點

Abstract

本論文的目的在於藉由無響室建造實體錄音環境，設計並且實現一個虛擬聆聽點系統。我們利用盲訊號分離(Blind Source Separation，BSS)、到達方向(Direction of Arrival，DOA)偵測、語音去雜訊等技術來建立虛擬聆聽位置的音訊，即虛擬聆聽點語音合成。為了達成這個目的，我們於自由聲場的無響室中佈置麥克風陣列，並且以揚聲器做為聲源錄製混合聲音訊號。 語音訊號合成主要可分成三個主要步驟，第一步驟是將錄製的混和訊號來估測原始各個音源訊號，此步驟一般使用盲訊號分離的技術來達成。第二步驟是為了估測原始各個音源訊號的來源方向，此步驟一般使用到達方向的技術來完成。第三步驟是為了在原始無麥克風錄音的虛擬位置上合成音訊，在我們系統中此步驟使用SLAB軟體來實現。 在實際的環境中，空氣中的雜訊干擾、音訊的失真是必然的。而這些因素會影響著訊號分離與到達方向的偵測。本篇論文中，我們提出許多主題來探討此兩種技術，我們提供許多數據來進行驗證分析。我們將數據分成三大部分：CASE-A、CASE-B.1和CASE-B.2。CASE-A為真實環境無響室中錄製的音源訊號。CASE-B.1是利用NASA研究中心所研發的SLAB軟體來錄製音源訊號。CASE-B.2為增加可加性白噪聲(Additive White Gaussian Noise，AWGN)於CASE-B.1而得的音源訊號。接著我們用語音去雜訊的技術，以改善人類主觀聽覺的品質，最後使用SLAB軟體來完成空間3D音訊處理程序。

The goal of this thesis is to design and implement a virtual listening point audio system by constructing a physical testing environment in an anechoic chamber. Several techniques are employed in implementing this system. They are blind source separation (BSS), direction of arrival (DOA) estimation and denoising filtering. The final outcome is constructing an audio signal at the desired virtual listening position, which is called Virtual Listening Point Audio Synthesis. In the Free Field Acoustic Room Chamber, each speaker represents a sound source and a microphone array records the received signals. The audio synthesis procedure can be divided into three major steps. The first step is to separate each source signal from the recorded mixed signals. This step is usually accomplished by using the blind source separation (BSS) technique. The second step is to estimate the direction (angle) of a sound source. This step is usually accomplished by using the direction of arrival (DOA) technique. The third step is to synthesize an audio signal at a virtual point, where the original recording microphone does not exist. In our system, this step is accomplished by using the SLAB software. In a real acoustic environment, noise and distortion are inevitable. They disturb the BSS performance and the DOA estimation. In this project, we study the effects of several key parameters in the system. We conduct experiments, collect data, and analyze data to verify the proposed schemes. The experiments are classified into CASE-A, CASE-B.1 and CASE-B.2. CASE-A denotes the speech source recorded from the microphone arrays in the anechoic chamber. CASE-B.1 denotes the signals produced by using SLAB developed by the NASA Ames Research Center to simulate the recorded mixture signals in an ideal acoustic environment. CASE-B.2 denotes that we add the Additive White Gaussian Noise (AWGN) to CASE-B.1. We also adopt audio denoising technique to improve the subjective hearing quality. Finally, the 3-D audios are synthesized with the aid of the SLAB software.